Machine Learning Pipeline For Power Electronics State-Of-Health Assessment And Remaining Useful Life Prediction

The rapid development of power electronics in modern systems, including electric vehicles, renewable energy grids, and industrial automation, has emphasized the need for effective monitoring and health management.Power electronic components, such as MOSFETs, are vulnerable to degradation over time due to various stress factors, including temperature cycles and electrical load.Accurately predicting the State-of-Health (SoH) and the Remaining Useful Life (RUL) of these devices is crucial for preventing unexpected failures, reducing downtime, and optimizing maintenance.This paper proposes a machine learning pipeline for SoH assessment and RUL prediction for power electronic devices.The system consists of two stages: a classification model to determine the health status and a regression model for RUL estimation.The classification stage utilizes Random Forest to predict the device's health status, while the RUL prediction is handled by a Bayesian Ridge Regression model.The proposed pipeline is tested on a real-world MOSFET dataset and achieves a classification accuracy of 84.3% and a Root Mean Square Percentage Error (RMSPE) of 1.8% for RUL prediction, demonstrating its effectiveness for real-time predictive maintenance applications.